Reproduce “From Visual Matching to Textual Conditioning: Cross-Domain Few-Shot Segmentation via Reference Inversion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiahao Nie, Yun Xing, Wenbin An, Qixin Zhang, Yap-Peng Tan, Alex C. Kot, Shijian Lu
RI-SAM3 adapts SAM3 for cross-domain few-shot segmentation by turning annotated reference images into reusable textual category conditions, rather than matching each target to a reference. It learns word-like and concept tokens and combines them across decoder layers, with confidence and presence supervision. In 1-shot Chest X-Ray it reaches 90.5% mIoU versus APSeg’s 84.1%; it is not best in DeepGlobe 5-shot.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.